Annual Journal of Hydraulic Engineering, JSCE, Vol.54, 2010, February
Annual Journal of Hydraulic Engineering, JSCE, Vol.54, 2010, February
SIMULATING OF LAND SURFACE SOIL
MOISTURE AND ENERGY FLUX IN NORTHERN
AFRICA USING A LAND DATA ASSIMILATION
SYSTEM AND UKMO OUTPUT
Hui LU1, Toshio KOIKE2, Kun YANG3, Xin LI4, Mohamed RASMY5, Hiroyuki TSUTSUI1,
Souhail BOUSSETTA1 and Katsunori TAMAGAWA6
1Member of JSCE, Ph.D., Researcher, Dept. of Civil Eng., Univ. of Tokyo (Bunkyo-ku, Tokyo 113-8656, Japan)
2Member of JSCE, Dr. Eng., Professor, Dept. of Civil Eng., Univ. of Tokyo (Bunkyo-ku, Tokyo 113-8656, Japan)
3Member of JSCE, Ph.D., Professor, Inst. of Tibetan Plateau Research, CAS (Beijing 100085, China)
4Ph.D., Professor, Cold and Arid Regions Envi. and Eng. Research Inst., CAS (Lanzhou 730000, China)
5Student Member of JSCE. M. Eng., Dept. of Civil Eng., University of Tokyo (Bunkyo-ku, Tokyo, 113-8656, Japan)
6Member of JSCE, M. Eng., Dept. of Civil Eng., University of Tokyo (Bunkyo-ku, Tokyo, 113-8656, Japan)
In this study, we generated hourly surface soil moisture and land surface energy fluxes in the
Northern Africa region, by using the land data assimilation system developed at the University of Tokyo
(LDAS-UT) which is driven by the UK Met Office (UKMO) weather prediction model output. The soil
moisture simulated by LDAS-UT was compared against that simulated by land surface model (LSM),
along with the 3-hourly Tropical Rainfall Measuring Mission (TRMM) rainfall in the Medjerda River
Basin. The comparison demonstrates the capability of LDAS-UT to partly decrease the error inherited
from biased forcing data. For the whole domain comparison, results also show that the soil moisture field
simulated by LDAS-UT is more realistic than that by LSM. The comparison between LDAS-UT soil
moisture with the Advanced Microwave Scanning Radiometers for EOS (AMSR-E) soil moisture
products indicates that LDAS-UT is able to provide more comprehensive information in the region and/or
during the period that satellite fails to provide observation. Moreover, by comparing Bowen ratio field
generated by LDAS-UT and by LSM with the land cover map, the capability of LDAS-UT to simulate
energy flux was verified. This study reveals the potential of LDAS-UT for simulating land surface fluxes
in ungauged regions.
Key Words: Land Data Assimilation System, UKMO, AMSR-E, Soil Moisture, Energy Fluxes, Bowen
Ratio
1. INTRODUCTION
Surface soil moisture is an essential variable
which governs the interactions between land surface
and the atmosphere1,2). In researches related to the
global warming and climate change, soil moisture is
serving as an excellent environmental indicator. The
distribution pattern of soil moisture, both spatial and
temporal, is the key to understanding the spatial
variability and scale problems. Moreover, soil
moisture also is an important factor in animal and
plant productivity and it can even be related to the
pattern of settlement in arid and semiarid regions.
Soil moisture profile can be observed at point
scale by using gravimetric methods or Time Domain
Reflector (TDR). These methods are commonly
used to provide accurate and temporal-continuous
soil moisture information and adopted by the
meteorology, hydrology and agriculture stations.
But these point information are not enough for the
regional research and application, and are also not
available in the far regions where difficult to access
and to build and maintain such stations. On the other
hand, Satellite remote sensing offers a possibility to
measure surface soil moisture at regional,
continental and even global scales. Passive
microwave remote sensing provides a means of
direct measurement of surface soil moisture3,4), with
coarse resolution (~ order of 10 km) and frequent
temporal coverage (daily or bi-daily), which can
partly satisfy the temporal resolution needed in the
meteorological modeling. However, in the field of
- 61 -
weather forecast and hydrology modeling, finer
temporal resolution would obviously improve the
accuracy and reliability of the forecast.
On the contrary to the limited observations
provided by in situ measurements and satellite
remote sensing, Land surface models (LSMs) 5,6) are
able to provide continuous estimating of surface soil
moisture in any scales. However, due to the model
initialization, parameter and forcing errors, and
inadequate model physics and/or resolution, quality
of the model predictions are usually not so good.
The Land Data Assimilation System (LDAS),
developed by merging observation information (in
situ measurements, satellites remote sensing and so
on) into dynamic models (i.e. LSMs), is expected to
provide high quality surface energy and water flux
estimates with adequate coverage and resolution7).
For the ungauged and poor-gauged regions, LDAS
is a reliable method to downscale the numerical
weather prediction model output and to generate
land surface status.
The climate in the Northern Africa varies from
humid equatorial regimes, through seasonally-arid
tropical regimes, to sub-tropical Mediterranean-type
regimes. All these climates exhibit differing degrees
of temporal variability, particularly with regard to
rainfall. Understanding the land atmosphere
interaction in this region, especially in the Sahel
region, has become the major challenge facing
African and African-specialist climate scientists in
recent years8,9). Simulating land surface fluxes over
this region is also a challenging topic, since in-situ
observations are not available for current research.
This study aims to deal with following two issues:
one is to generate land surface fluxes over this
region by using LDAS-UT which is driven by the
weather model output data; the other is to
investigate the capability of LDAS-UT by
comparing the its results with LSM simulations and
satellite products, such as 3-hourly TRMM rainfall
data10) and AMSR-E soil moisture products.
In the following sections, we briefly introduce the
components and algorithm of LDAS-UT11). The
condition of application region and the data used in
this study will be introduced in section 3. The
simulated fluxes by LDAS-UT and LSM will be
shown in section 4 with some discussion and
analysis. This paper ends at section 5 with some
conclusions.
2. THE LAND
SYSTEM
DATA
ASSIMILATION
LDAS-UT is a variational data assimilation
system. The brightness temperature data observed
by AMSR-E at C-band and X-band are assimilated
into the system to get better soil moisture estimation
and then to improve the land surface energy fluxes
simulation.
2.1 Three Components of LDAS-UT
From the view of system structure, LDAS-UT is
composed by three components: a dynamic model
for propagating status variables, an observation
operator for introducing new information into the
dynamic model, and an optimization model to
minimize the cost function of the system.
The dynamic model of LDAS-UT is a Land
Surface Model (LSM) named Simple Biosphere
model (SiB2)5), which is used to calculate surface
fluxes and soil moisture for each time step.
The model operator of LDAS-UT is a radiative
transfer model (RTM), which estimate microwave
brightness temperature from corresponding land
surface status. The RTM used in LDASUT has two
components: a volume scattering part and a surface
scattering part12). The volume scattering part
simulates the radiative transfer process inside soil
layer by a 4-stream based RTM in which the
multiply scattering effects of dry soil medium is
calculated by the dense media radiative transfer
model (DMRT)13). The surface scattering part
simulates the surface scattering effects at the landatmosphere interface by Advanced Integral
Equation Method (AIEM)14).
The minimization scheme of LDAS-UT is the
shuffled complex evolution method15), which search
for optimal values of soil moisture through
minimizing the difference between modeled and
observed brightness temperature.
2.2 Parameters Optimization by LDAS-UT
LDAS-UT employs a dual-pass assimilation
technique. In pass 1, the model parameters are
estimated by using long-term (~ months)
meteorological forcing data and remote sensed
brightness temperature data, while the soil moisture
and fluxes are assimilated in pass 2. More details of
the algorithm can be found in the figure 1 of Yang
et al.11).
The initial parameters of this system are obtained
from global data sets, for example, the LAI from
Moderate Resolution Imaging Spectroradiometer
(MODIS), and the soil and vegetation parameters
and land cover types from International Satellite
Land-Surface Climatology Project (ISLSCP)
initiative II16).
As we know, the parameters provided by the
global data set sometimes are problematic, due to
the different spatial resolutions and the original
application purposes of various data sources. The
pass 1 of LDAS-UT, parameter optimization pass, is
able to generate a good parameter set from the
global initial parameters. The physical basis of the
parameter optimization is following: (1) the
simulated soil moisture is mainly dependent on the
- 62 -
forcing data and soil parameters; (2) the parameters
are almost stable values during a season (around 3
months); (3) the satellite provides reliable
observation. Based on those assumptions, we run
the SiB2 continuously for a period around two
months, which is so-called optimization window.
During this optimization window, a series of
brightness temperature was estimated with RTM
when the AMSR-E observation are available. The
summation of the squared difference between the
simulated TB and AMSR-E observed TB is the socalled cost. Through minimizing the cost, the
parameter set are optimized. The optimized
parameters in pass 1 are soil porosity, soil texture
(percentage of sand and clay), surface roughness
parameters (r.m.s height and correlation length),
vegetation RTM parameters (χ and b′). Besides
these parameters, in order to get better performance,
the initial soil moisture and temperature at each
layer are also optimized in the first run.
3. APPLICATION REGION AND USED
DATA
3.1 Description to The Application Region
Northern Africa, experiencing a dramatic climate
regimes variation from the semiarid desert in the
south to a mild Mediterranean climate in the North,
has an environment vulnerable to the climate
change. On the other hand, it is very difficult to
monitor the environmental changes in this region
due to its severe climate conditions.
In this study, the application region is mainly
located at Tunis, from 7E to 12 E and from 30N to
37.5 N. The application period is from beginning of
January to the end of March, 2003. In this season,
some heavy rainfall usually occurred and caused
flood disaster.
3.2 Used AMSR-E Data Set
The brightness temperature of AMSR-E observed
at 6.925 GHz and 18.7GHz were used as the
observation data in this research, by considering the
Radio Frequency Interference (RFI) effect is not so
strong in this region. Observation times of AMSR-E
are local noon and midnight. In this application,
only the midnight observation data was used since it
was obtained at nighttime overpass and the land
surface temperature profile was almost uniform. The
assimilation window of LDAS-UT is, therefore,
decided as one day. The resolution of LDAS-UT is
also decided by the resolution of AMSR-E data, as
the 0.5 degree.
3.3 Used Meteorological Forcing Data
In this research, both the LDAS-UT and LSM
(SiB2) were driven by UKMO numerical weather
prediction model output to generate hourly land
surface fluxes. The UKMO data set was archived at
the model output data base of the Coordinated
Energy and Water Cycle Observation Project
(CEOP) 17). The original data is offered in a 1.25
degree resolution four times per day (at 03:00,
09:00, 15:00 and 21:00 UTC). Interpolation was
used to get hourly forcing data.
4. RESULTS AND DISCUSSIONS
With UKMO model output as the meteorological
forcing data, we run LDAS-UT and LSM separately
to get two sets of land surface fluxes. Since there are
not in-situ observed fluxes available, we investigate
the simulated results from three aspects: (1)
comparing hourly soil moisture time series of
LDAS-UT and LSM with the 3-hourly TRMM
rainfall in a basin scale; (2) comparing snapshot of
simulated soil moisture with AMSR-E products and
topographical map over whole domain; (3)
comparing monthly averaged Bowen ratio of
LDAS-UT and LSM with the satellite land cover
map.
4.1 Comparing hourly soil moisture time
series in the Medjerda river basin
As the first step, we focused the comparison in
the Medjerda river basin (35.9N – 36.6N, 8.3E –
10.0E), the largest water resource in Tunisia, in a
time scale of a rainfall event.
Fig. 1 The location of Medjerda river basin
Figure 1 shows a geography map of the location
of Medjerda basin. The Medjerda River flows
through the northern part of Tunisia to the
Mediterranean Sea and creates a catchment of
23,700km2. It has a population of 2,100,000 within
its basin. A large-scale flood occurred in the
Medjerda River Basin in January 2003, which
flooded the lower plain area for a month and caused
severe socio-economic damage.
The hourly soil moisture estimation field
generated by LDAS-UT and LSM were averaged
over whole basin and were shown in figure 2 as the
- 63 -
From figure 2, it is clear that TRMM detected a
heavy rainfall event started around 13:00 of January
10, 2003, while UKMO provided a delayed rainfall
starting from 0:00 of 11th. Since UKMO rainfall
data is biased during January 10th, the soil moisture
simulated by LSM only did not show any increase
during this day. On the contrary, since the land wet
situation was detected by satellite and this
information was assimilated into LDAS-UT, the soil
moisture simulated by LDAS-UT increased. By
assimilating AMSR-E brightness temperature into
LSM, LDAS-UT therefore is able to partly decrease
the error caused by biased forcing data.
0.5
5
Soil Moiture
0
0.4
10
0.3
15
0.2
TRMM_Rain
LDAS
0.1
Rainfall (mm)
Hourly Soil Moisture over Medjerda Basin
0.6
20
UKMO_Rain
LSM
25
0
1/10
1/11
1/12
Fig. 2 Averaged hourly soil moisture generated by LDAS-UT
and LSM in the Medjerda river basin, from January 10 to 12,
2003. Averaged TRMM rainfall and UKMO rainfall were also
plotted as vertical bars.
solid line and dash line, respectively. The 3-hourly
rainfall from TRMM and UKMO rainfall were
plotted as the vertical blank bars and filled bars.
4.2 Comparing soil moisture over whole
domain
Figure 3 shows the snapshots of three sets of soil
Fig. 3 Comparison of soil moisture field of LDAS-UT (upper row), LSM (middle row) and AMSR-E (bottom row) over whole study
domain, from left to right in the order of Jan 9, 10, 11 and 12, 2003.
- 64 -
(a)
(a)
(b)
Fig. 5 Comparison the monthly averaged Bowen Ratio
simulated by LDAS-UT (a) with that by LSM (b), over whole
study region.
(b)
Fig. 4 (a) Topographical map and (b) land cover map of the
study region.
moisture over the whole domain: Upper row and
middle row are the soil moisture of LDAS-UT and
LSM at the 00:00 of January 9, 10, 11 and 12, 2003,
respectively. The bottom row is the AMSR-E soil
moisture products observed at midnight.
As shown by the LDAS-UT soil moisture map in
figure 3, the southwest of study region, where part
of Sahara desert located, is the driest part. LDASUT also estimated high soil moisture value in the
Northwest region, where is the edge of
Mediterranean Sea. Comparing the soil moisture
patterns with the topographical map (figure 4a), it is
clear that the soil moisture simulated by LDAS-UT
is more realistic than that of LSM, which shows less
variation and poor relationship to the topography.
For instance, LDAS-UT simulated a wet region
around (34N, 8.5E), where is the salt lake named
Chott el Djerid, while LSM failed to generate a wet
spot here. This wet region was also detected by
AMSR-E. But we also found that AMSR-E failed to
provide full coverage over whole domain, partly due
to the limitation of algorithm and sensors. One more
thing which should be noted is that AMSR-E can
only provide twice observation per day while
LDAS-UT is able to generate hourly soil moisture.
Comparing to the AMSR-E soil moisture products,
LDAS-UT therefore is able to provide more
information of spatial and temporal variation in
surface soil moisture.
4.3 Comparing Bowen Ratio over whole
domain
Besides soil moisture, LDAS-UT also generated
the land surface energy components: net radiation,
sensible heat, latent heat and grand soil heat. As we
know, the land surface energy flux is important to
understand the land atmosphere interaction and also
to weather forecast and climate study. But due to the
lack of in situ observation data, we only can validate
the LDAS-UT energy flux output indirectly.
Figure 5 shows the monthly averaged Bowen
Ration of LDAS-UT (5a) and that of LSM (5b).
From figure 5, it is clear that, for both simulations,
the sensible heat flux generally dominates the
energy budget over whole domain, except the north
coast region where the latent heat flux is
dominative. The highest Bowen ratio was found in
the southwest of study region, Sahara desert, again.
By comparing the Bowen ratio show in Figure 5, it
is found that LDAS-UT produces more latent heat
dominative regions than LSM does. LSM fails to
generate the latent heat dominative region around
the west coast line and the Chott el Djerid lake.
Referring to the land cover map (figure 4b), the
distribution patterns of LDAS-UT Bowen ratio are
more realistic than that of LSM.
5. CONCLUSIONS
Regional soil moisture distribution is crucial for
studies of agricultural, hydrological, ecological and
atmospheric processes. In this study, we generated
hourly soil moisture and energy fluxes distribution
in the Northern Africa region, with using LDAS-UT
forced by UKMO numerical weather prediction
model output.
The hourly soil moisture time series of LDAS-UT
outputs were compared with those of LSM, during a
heavy rainfall event in the Medjerda river basin. The
results show that LDAS-UT can partly decrease the
errors inherited from biased forcing data. Such
advantages are based on the assimilating of AMSRE TB data, which directly related to the land surface
status.
For whole study region, LDAS-UT soil moisture
shows a good agreement with the topographical
map, giving a more realistic pattern than that of
LSM. The comparison between LDAS-UT soil
- 65 -
moisture and AMSR-E products demonstrates that
LDAS-UT is able to provide more comprehensive
information in the region and/or during the period
that satellite fails to provide observation
Moreover, the Bowen ratio distribution generated
by LDAS-UT was also represented in this study. By
comparing the results to the geographical and land
cover map, we can conclude that the Bowen ratio
distribution estimated by LDAS-UT is more realistic
than that generated by LSM only.
Due to the lack of in situ observation data,
quantitative validation is impossible now. The
qualitative results from this study can be served as
the primary step for the land atmosphere interaction
research in this region. Based on the agreement of
GEOSS (the Global Earth Observation System of
Systems) African Water Cycle Symposium
(AWCS), in situ observation network will be
organized in near future. With the in situ
observation data provided by AWCS, LDAS-UT
can be well calibrated and its performance can be
further improved.
In this study, all the data and parameters were
obtained from global data set: meteorological
forcing data from UKMO output; satellite remote
sensing data from AMSR-E; vegetation data from
MODIS and parameters from ISLSCP II. All data
are public accessible and free. Therefore, The
experiences earned from this study are also useful
for the region where is ungauged or poor-gauged.
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
ACKNOWLEDGMENTS:
This study was carried out as a part of Coordinated
Enhanced Observing Period (CEOP) project funded
by the Japan Science and Technology Agency and
the Special Coordination Funds for Promoting
Science and Technology. The authors express their
so deep gratitude to them. Furthermore we would
like to thank the Japan Aerospace Exploration
Agency (JAXA) and UKMO for their support in
providing the necessary data set.
13)
14)
15)
REFERENCES
16)
1)
17)
2)
Entekhabi, D., I. Rodrigues-Iturbe, and F. Castelli,:
Mutual interaction of soil moisture state and atmospheric
processes, Journal of Hydrology, 184, 3-17, 1996.
Betts AK, Ball JH, Beljaars ACM et al. : The land
surface-atmosphere interaction: a review based on
observational and global modeling perspectives. Journal
of Geophysical Research, 101, 7209–7225, 1996.
- 66 -
Njoku, E.G. and Entekhabi, D., : Passive microwave
remote sensing of soil moisture, Journal of Hydrology
184, 101-129, 1996
Njoku, E.G., Jackson, T.J., Lakshmi, V., Chan, T.K. and
Nghiem, S.V.,: Soil moisture retrieval from AMSR-E,
IEEE Transactions on Geoscience and Remote Sensing 41,
215-229, 2003.
Sellers PJ, Randall DA, Collatz GJ et al.: A revised land
surface parameterization (SiB2) for atmospheric GCMs.
Part I: model formulation. Journal of Climate, 9, 676–705,
1996.
Gao, Z., N. Chae, J. Kim, J. Hong, T. Choi, and H. Lee:
Modeling of surface energy partitioning, surface
temperature, and soil wetness in the Tibetan prairie using
the Simple Biosphere Model 2 (SiB2), Journal of
Geophysical
Research,
109,
D06102,
doi:10.1029/2003JD004089, 2004.
Reichle, R.H. and Koster, R.D.: Global assimilation of
satellite surface soil moisture retrievals into the NASA
catchment land surface model. Geophysics research
letters, 32, L02404, doi:10.1029/2004GL021700, 2005.
Xue,Y. and Shukla,J.: Model simulation of the influence
of global SST anomalies on Sahel rainfall, Monthly
Weather Review., 126, 2782-2792, 1998.
Cunnington,W.M. and Rowntree,P.R.: Simulations of the
Saharan atmosphere: dependence on moisture and albedo
Quarterly Journal of the Royal Meteorological Society,
112, 971-999, 1996
Huffman, G.J., R.F. Adler, D.T. Bolvin, G. Gu, E.J.
Nelkin, K.P. Bowman, E.F. Stocker, D.B. Wolff: The
TRMM multi-satellite precipitation analysis: quasi-Global,
multi-Year, combined-sensor precipitation estimates at
fine scale. Journal of Hydrometeorology, 8, 33-55, 2007.
Yang, K., Watenabe, T., Koike, T., Li, X., Fujii, H.,
Tamagawa, H., Ma, Y. and Ishizawa H.: An autocalibration system to assimilate AMSR-E data into a land
surface model for estimating soil moisture and surface
energy budget, Journal of the Meteorological Society of
Japan, 85A, 229-242, 2007.
Lu, H., Koike, T., Fujii, H., Hirose, N. and Tamagawa, K.:
A radiative transfer model and an algorithm for soil
moisture including very dry conditions, In Proceedings of
the IEEE 2005 International Geoscience and Remote
Sensing Symposium (IGARSS’05) II, 1131 – 1134, 2005
Wen, B, L. Tsang, D. P. Winebrenner, and A. Ishimura:
Dense media radiative transfer theory: comparison with
experiment and application to microwave remote sensing
and polarimetry, IEEE Transactions on Geoscience and
Remote Sensing, 28, 46-59, 1990.
Chen, K.S., T. D. Wu, L. Tsang, Q. Li, J. C. Shi, and A.
K. Fung: Emission of rough surfaces calculated by the
integral equation method with comparison to threedimensional moment method Simulations, IEEE
Transactions on Geoscience and Remote Sensing, 41, pp.
90-101, 2003
Duan, Q., V.K. Gupta, and S.Sorooshian: A shuffled
complex evolution approach for effective and efficient
global minimization, Journal of Optimization Theory and
Applications, 76, 501-521,1993
Forrest, G.H., Collatz, G., Los, S., Colstoun, E.B., Landis,
D.: ISLSCP Initiative II, NASA, DVD/CD-ROM, 2005.
Koike, T.: The Coordinated Enhanced Observing Period –
an initial step for integrated global water cycle
observation, WMO Bulletin, 53(2), 1-8, 2004.
(Received September 30, 2009)